Re: [PATCH v4 1/3] ethdev: introduce maximum Rx buffer size

[Ferruh Yigit <ferruh.yigit@amd.com>]

On 11/3/2023 2:21 AM, lihuisong (C) wrote:
> 
> 在 2023/11/3 0:35, Ferruh Yigit 写道:
>> On 11/2/2023 12:16 PM, Huisong Li wrote:
>>> The "min_rx_bufsize" in struct rte_eth_dev_info stands for the minimum
>>> Rx buffer size supported by hardware. Actually, some engines also have
>>> the maximum Rx buffer specification, like, hns3, i40e and so on. If mbuf
>>> data room size in mempool is greater then the maximum Rx buffer size
>>> per descriptor supported by HW, the data size application used in each
>>> mbuf is just as much as the maximum Rx buffer size instead of the whole
>>> data room size.
>>>
>>> So introduce maximum Rx buffer size which is not enforced just to
>>> report user to avoid memory waste. In addition, fix the comment for
>>> the "min_rx_bufsize" to make it be more specific.
>>>
>>> Signed-off-by: Huisong Li <lihuisong@huawei.com>
>>> Acked-by: Chengwen Feng <fengchengwen@huawei.com>
>>> Acked-by: Morten Brørup <mb@smartsharesystems.com>
>>> ---
>>>   doc/guides/rel_notes/release_23_11.rst |  7 +++++++
>>>   lib/ethdev/rte_ethdev.c                |  7 +++++++
>>>   lib/ethdev/rte_ethdev.h                | 10 +++++++++-
>>>   3 files changed, 23 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/doc/guides/rel_notes/release_23_11.rst
>>> b/doc/guides/rel_notes/release_23_11.rst
>>> index 95db98d098..d4f7d5b266 100644
>>> --- a/doc/guides/rel_notes/release_23_11.rst
>>> +++ b/doc/guides/rel_notes/release_23_11.rst
>>> @@ -122,6 +122,13 @@ New Features
>>>     a group's miss actions, which are the actions to be performed on
>>> packets
>>>     that didn't match any of the flow rules in the group.
>>>   +* **Added maximum Rx buffer size to report.**
>>> +
>>> +  Introduced the ``max_rx_bufsize`` field representing the maximum Rx
>>> +  buffer size per descriptor supported by HW in structure
>>> ``rte_eth_dev_info``
>>> +  to report user and to avoid wasting space of mempool.
>>> +  Its value is UINT32_MAX if driver doesn't report it.
>>> +
>>>   * **Updated Amazon ena (Elastic Network Adapter) net driver.**
>>>       * Upgraded ENA HAL to latest version.
>>> diff --git a/lib/ethdev/rte_ethdev.c b/lib/ethdev/rte_ethdev.c
>>> index af23ac0ad0..03539bb6c3 100644
>>> --- a/lib/ethdev/rte_ethdev.c
>>> +++ b/lib/ethdev/rte_ethdev.c
>>> @@ -2112,6 +2112,7 @@ rte_eth_rx_queue_setup(uint16_t port_id,
>>> uint16_t rx_queue_id,
>>>       struct rte_eth_dev *dev;
>>>       struct rte_eth_dev_info dev_info;
>>>       struct rte_eth_rxconf local_conf;
>>> +    uint32_t vld_bufsize;
>>>   
>> I guess 'vld' is valid, but that is not accurate. Can we rename it
>> something like 'buf_data_size'?
> Ack
>>
>>>       RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV);
>>>       dev = &rte_eth_devices[port_id];
>>> @@ -2158,6 +2159,11 @@ rte_eth_rx_queue_setup(uint16_t port_id,
>>> uint16_t rx_queue_id,
>>>               return ret;
>>>             mbp_buf_size = rte_pktmbuf_data_room_size(mp);
>>> +        vld_bufsize = mbp_buf_size - RTE_PKTMBUF_HEADROOM;
>>> +        if (vld_bufsize > dev_info.max_rx_bufsize)
>>> +            RTE_ETHDEV_LOG(INFO,
>>> +                "For port_id=%u the max data size per mbuf in Rx is
>>> %u instead of the whole data room(%u).\n",
>>> +                port_id, dev_info.max_rx_bufsize, vld_bufsize);
>>>
>> This log gives some information, but it is not clear what user should
>> do. My concern is this can confuse end users (not developers but users
>> of dpdk application).
>>
>> Since audience of this message is DPDK application developer, most of
>> the times end user don't have anything to do with this log, what do you
>> think to change log level to 'debug'?
> Ack
>>
>> Also can we add some action to it, something like:
>> "mbuf data buffer size is bigger than what device can utilize, consider
>> reducing mbuf size."
> What do you think of this log, Ferruh?
> "The mbuf data buffer size (%u) is bigger than the max size (%u) device
> with port_id=%u can utilize, consider reducing mbuf size.\n",
>

Can move port_id to the beginning, and can make wording passive, what about:

"For port_id=%u, the mbuf data buffer size (%u) is bigger than max
buffer size (%u) device can utilize, mbuf size can be reduced.\n"

>>
>>
>>
>>>       } else if (rx_conf != NULL && rx_conf->rx_nseg > 0) {
>>>           const struct rte_eth_rxseg_split *rx_seg;
>>>           uint16_t n_seg;
>>> @@ -3757,6 +3763,7 @@ rte_eth_dev_info_get(uint16_t port_id, struct
>>> rte_eth_dev_info *dev_info)
>>>       dev_info->min_mtu = RTE_ETHER_MIN_LEN - RTE_ETHER_HDR_LEN -
>>>           RTE_ETHER_CRC_LEN;
>>>       dev_info->max_mtu = UINT16_MAX;
>>> +    dev_info->max_rx_bufsize = UINT32_MAX;
>>>         if (*dev->dev_ops->dev_infos_get == NULL)
>>>           return -ENOTSUP;
>>> diff --git a/lib/ethdev/rte_ethdev.h b/lib/ethdev/rte_ethdev.h
>>> index a53dd5a1ef..7133b73d26 100644
>>> --- a/lib/ethdev/rte_ethdev.h
>>> +++ b/lib/ethdev/rte_ethdev.h
>>> @@ -1723,7 +1723,15 @@ struct rte_eth_dev_info {
>>>       uint16_t min_mtu;    /**< Minimum MTU allowed */
>>>       uint16_t max_mtu;    /**< Maximum MTU allowed */
>>>       const uint32_t *dev_flags; /**< Device flags */
>>> -    uint32_t min_rx_bufsize; /**< Minimum size of Rx buffer. */
>>> +    /** Minimum Rx buffer size per descriptor supported by HW. */
>>> +    uint32_t min_rx_bufsize;
>>> +    /**
>>> +     * Maximum Rx buffer size per descriptor supported by HW.
>>> +     * The value is not enforced, information only to application to
>>> +     * optimize mbuf size. Its value is UINT32_MAX when not specified
>>> +     * by the driver.
>>> +     */
>>> +    uint32_t max_rx_bufsize;
>>>       uint32_t max_rx_pktlen; /**< Maximum configurable length of Rx
>>> pkt. */
>>>       /** Maximum configurable size of LRO aggregated packet. */
>>>       uint32_t max_lro_pkt_size;
>> .